cuda12_2-2.8.0 container unable to detect gpu #216
Comments
This says that the system can't recognise your CUDA install. Are you running CUDA 12.2+ locally, and did you start the container using - |
|
i used docker compose: version: "3.8"
services:
codeprojectai:
image: codeproject/ai-server:cuda12_2-2.8.0
container_name: CodeProjectAI
restart: unless-stopped
ports:
- "32168:32168"
deploy:
resources:
reservations:
devices:
- driver: nvidia
capabilities: [gpu]
volumes:
- /etc/codeproject/ai:/etc/codeproject/ai
- /opt/codeproject/ai:/app/modules
environment:
- NVIDIA_VISIBLE_DEVICES=all
- NVIDIA_DRIVER_CAPABILITIES=allshould i have used docker with --gpus all instead? i got this installed on the host and frigate sees the same which uses the same flags on its docker compose: |
|
I just realised you are running an older docker container. Try using the latest version: '3.9'
services:
CodeProjectAI:
image: codeproject/ai-server:cuda12_2
container_name: "codeproject-ai-server-cuda"
restart: unless-stopped
ports:
- "32168:32168/tcp"
- "32168:32168/udp"
environment:
...If that doesn't work it might be worth trying: |
|
Will try later.Thanks for pointing that out, kinda new to containers. On Dec 13, 2024, at 1:43 AM, Chris Maunder ***@***.***> wrote:
I just realised you are running an older docker container. Try using the latest
version: '3.9'
services:
CodeProjectAI:
image: codeproject/ai-server:cuda12_2
container_name: "codeproject-ai-server-cuda"
restart: unless-stopped
ports:
- "32168:32168/tcp"
- "32168:32168/udp"
environment:
...
If that doesn't work it might be worth trying:
docker run --name CodeProject.AI -d -p 32168:32168 --gpus all ^
--mount type=bind,source=C:\ProgramData\CodeProject\AI\docker\data,target=/etc/codeproject/ai ^
--mount type=bind,source=C:\ProgramData\CodeProject\AI\docker\modules,target=/app/modules ^
codeproject/ai-server:cuda12_2
—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>
|
|
Did you install the nvidia container toolkit on the host? I just went thru this exercise with Debian 12. Have you tried one of the test containers that run nvidia-smi in a container, just to see that docker is passing it thru correctly? |
|
Yes, got container toolkit installed and working. Nvidia smi installed on frigate and working as well.On Dec 15, 2024, at 9:37 AM, Randell Hodges ***@***.***> wrote:
Did you install the nvidia container toolkit on the host? I just went thru this exercise with Debian 12. Have you tried one of the test containers that run nvidia-smi in a container, just to see that docker is passing it thru correctly?
—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>
|
|
I am using codeproject/ai-server:cuda12_2-2.9.5 if that might make a difference. My setup. I have a Debian 12 VM running in proxmox. I use an Nvidia P4 and pass a VGPU to the Debian 12 VM. In that VM, I installed the vgpu drivers, the license, nvidia container toolkit, docker and that's about it. My docker compose looks almost identical except the that I'm using 2.9.5. I am able to docker exec -it container bash and then poke around and run nvidia-smi inside the container and see my card and everything is working. It sounds like you have other docker containers on the same host using the card. So it sounds like the host is fine. I've never tried docker with a consumer card and docker. Is there any issues sharing a consumer card to multiple containers at the same time? I'd give 2.9.5 a try. |
Area of Concern
Describe the bug
i'm using the codeproject/ai-server:cuda12_2-2.8.0
cuda is only used on the yolo5 model and not by the facial recognition model as well.
i just started with the project, so am not quite sure if i misconfigured anything so far.
Expected behavior
cpai server uses cuda for all modules in the container, not just object detection
Screenshots
If applicable, add screenshots to help explain your problem.
Your System (please complete the following information):
Additional context
docker logs, docker compose, and nvidia smi
opt/codeproject$ docker logs CodeProjectAI
The text was updated successfully, but these errors were encountered: